IMA corner case with forward declarations of statics

[Richard Sandiford]

Richard Henderson <rth@redhat.com> writes:
> The point is that we *already* have a situation where we get broken
> code at -fno-unit-at-a-time, and thus we should diagnose this error
> all of the time, regardless of how often users have traditionally
> made this mistake and gotten away with it.

For the record, I don't think this isn't really true for MIPS
(which was the example given upthread).  We disable use of
explicit relocs in PIC when -fno-unit-at-a-time.

Richard